Detection of geometric structures in an image by Helmholtz principle

Agnes Desolneux
CNRS, France
MAP5, Universite Paris5

The so-called Helmholtz principle is a principle of visual perception which roughly says that,
in an image, the structures we see are the ones which have a very low probability to appear
in noise, and which belong to a given class of pre-defined geometric objects (namely the
class of "gestalts" defined by the gestalt school of psychophysiology). We will explain how
a mathematical model based on this Helmholtz principle can be built, and how it can be efficently
applied to the problem of the detection of geometric structures in an image: alignments, edges,
smooth curves, groups of objects which have the same orientation (or color, or diameter),
clusters, vanishing points.

Back to Multiscale Geometric Analysis: Theory, Tools, and Applications